plt_craft<-ggplot(data=AFSNT %>% 
                    mutate(DLY=ifelse(DLY=="Y",1,0),
                           DLY_B=ifelse(str_detect(DRR,"B"),1,0)) %>% 
                    group_by(ARP) %>% 
                    summarise(a=sum(DLY_B)) ,aes(x=ARP,y=sort(a, decreasing = T)))+
  geom_col(aes(fill=ARP),alpha=0.6)+ scale_fill_manual(values=my_palette)+
  ggtitle("공항 사유로 인한 지연 수")+
  xlab("공항")+
  ylab("지연 수")+
  theme(plot.title = element_text(size = 20,hjust=0.5),
        legend.position = "none", axis.text.x = element_text(angle=45, hjust = 1))
plt_craft

# 예측해야 하는 DLY 와 공통인 항공편만 정리 
afsnt_common <- afsnt%>% filter(FLT %in% AFSNT_DLY$FLT) %>% arrange(DATE,FLT,SCT) 

REG_FLT <- afsnt_common  %>% filter(AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)

# 항공사 별로 운항 수가 가장 많은 Top REG 뽑기 
A_REG_FLT <- afsnt_common  %>% filter(FLO=='A',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)
B_REG_FLT <- afsnt_common  %>% filter(FLO=='B',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)
F_REG_FLT <- afsnt_common  %>% filter(FLO=='F',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)
H_REG_FLT <- afsnt_common  %>% filter(FLO=='H',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)
I_REG_FLT <- afsnt_common  %>% filter(FLO=='I',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)
J_REG_FLT <- afsnt_common  %>% filter(FLO=='J',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)
L_REG_FLT <- afsnt_common  %>% filter(FLO=='L',AOD =='D',DATE>='2019-03-31')%>% select(REG,FLT,DATE)

# REG 중 top 10 만 추리기 
A_TOPREG <- A_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
A_TOPREG <- A_TOPREG[1:10,]

B_REG_FLT <- B_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
B_REG_FLT <- B_REG_FLT[1:10,]

F_REG_FLT <- F_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
F_REG_FLT <- F_REG_FLT[1:10,]

H_REG_FLT <- H_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
H_REG_FLT <- H_REG_FLT[1:10,]

I_REG_FLT <- I_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
I_REG_FLT <- I_REG_FLT[1:10,]

J_REG_FLT <- J_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
J_REG_FLT <- J_REG_FLT[1:10,]

L_REG_FLT <- L_REG_FLT %>% group_by(REG) %>% summarise(n=n()) %>%
  arrange(desc(n))%>% as.data.frame() %>% select(REG)
L_REG_FLT <- L_REG_FLT[1:10,]

# REG 별로 연관 분석시행 
# REG 별  Shopper  = Day  //  Item = FLT

# A 항공사  
A_tran <- REG_FLT %>% filter(REG %in% A_TOPREG) #top 10 reg 있는 것만 선택
A_tran$DATE <- A_tran$DATE %>% as.character.Date()
A_unique_date <- unique(A_tran$DATE)# 날 씨 는 총 92 일 

# 빈 리스트 생성 by assign 함수 

y<-list(); A_top <-list()

# top10 FLT - 루프로 돌리기 
# REG별 &날짜별로 나열 후, 해당 REG가 다니는 FLT 를 세개 씩 묶어 정리한다. -> 서로의 연속성을 살리기위해
for (l in 1:10) {
  y<-list()
  for (i in seq(length(A_unique_date))){
    FLT_COL <- A_tran %>% filter(REG==A_TOPREG[l], DATE== A_unique_date[i])  %>% select(FLT) 
    FLT_COL <- unlist(FLT_COL) %>% as.vector()
    
    if(length(FLT_COL) != 0){
      if(length(FLT_COL) != 1){
        if(length(FLT_COL) != 2){
          len = length(FLT_COL)-2
          
          for (j in seq(len)){
            k = j+2
            x <- c(FLT_COL[j:k])
            y <- append(y,list(x))
          }
        }  
      } 
    }
  }
  A_top[[l]] <- y
}  

# Transaction data 로 변환 
# apiori 함수 사용
# https://m.blog.naver.com/leedk1110/220788082381

####변수에 대한 설명
# https://m.blog.naver.com/PostView.nhn?blogId=gkenq&logNo=10188110816&proxyReferer=https%3A%2F%2Fwww.google.com%2F

A_rule <- list()
for ( i in seq(10)){
  A_rule[[i]] <-  apriori(A_top[[i]],control = list(verbos=F), parameter = list(support=0.01,minlen=3))
}
# inspect(sort(A_rule[[7]],by='lift'))

# 얻어낸 rule 들을 data frame화 
# A 항공사 최다 빈도 REG 로 확인 해보기 
A_rule_data <- list()
for (i in 1){
  A_rule_data[[i]] <- inspect(sort(A_rule[[i]],by='lift')) %>%  as.data.frame()
}

# 1st REG
A_rule_data[[1]][1:5,] 

# MIST & VIS_MNM plot
weather_arp<-fread("./data/weather_arp.csv", stringsAsFactors=F)
plt_mist<-ggplot(data=weather_arp[weather_arp$SDT_MM==9&VIS_MNM!=1000,],aes(x=MIST,y=VIS_MNM))+
  geom_point(color="blue")+
  geom_smooth(method="lm",col="red",se=F)+
  ggtitle("Feature Engineering : MIST")+
  theme(plot.title = element_text(size = 20,hjust=0.5))
plt_mist

plt_HM_INV<-ggplot(data=weather_arp[weather_arp$SDT_MM==9&VIS_MNM!=1000,],aes(x=HM_INV,y=VIS_MNM))+
  geom_point(color="blue")+
  geom_smooth(method="lm",col="red",se=F)+
  ggtitle("Feature Engineering : HM_INV")+
  theme(plot.title = element_text(size = 20,hjust=0.5))
plt_HM_INV

AFSNT <- fread("./data/AFSNT.csv",stringsAsFactors=F,header=T)
afsnt<-AFSNT

arp_tuner <- function(df,i){
  afsnt_d <- df %>% filter(AOD=='D' & ARP==paste0('ARP',i))
  afsnt_a <- df %>% filter(AOD=='A' & ODP==paste0('ARP',i))
  afsnt_arp <- rbind(afsnt_d, afsnt_a)
  afsnt_arp <- afsnt_arp %>% select(SDT_YY,SDT_MM,SDT_DD,STT,ARP,ODP,AOD,FLT) %>%
    mutate(STTT=paste(SDT_YY,SDT_MM,SDT_DD,STT,sep=':'),
           POSIX=as.POSIXct(STTT,format='%Y:%m:%d:%H:%M')) %>% arrange(POSIX)
  afsnt_arp$schedule <- strptime(afsnt_arp$STTT,'%Y:%m:%d:%H:%M')
  schedule_time <- afsnt_arp$schedule$hour*60 + afsnt_arp$schedule$min
  count <- c(1); tmp <- 1; cnt <- 0
  for (i in 2:nrow(afsnt_arp)){
    if(afsnt_arp$schedule[i]$mday!=afsnt_arp$schedule[i-1]$mday){
      tmp <- i; count[i] <- 1
    } else{
      cnt <- 0
      for (j in tmp:i){
        if(schedule_time[i] - schedule_time[j] <= 60){
          cnt <- cnt+1
        }
      }
      count[i] <- cnt
    }
  }
  afsnt_arp <- cbind(afsnt_arp, count) %>% select(-c('POSIX','STTT','schedule'))
  return(afsnt_arp)
}

complexity <- arp_tuner(afsnt,1)
for (i in 2:15){
  complexity <- rbind(complexity, arp_tuner(afsnt,i))
}

AFSNT <- fread("./data/AFSNT.csv",stringsAsFactors = F,header=T)
afsnt<-AFSNT
## AOD가 D일 때
arp_tuner_d <- function(df,i){
  afsnt_d <- df %>% filter(AOD=='D' & ARP==paste0('ARP',i))
  #afsnt_a <- df %>% filter(AOD=='A' & ARP==paste0('ARP',i))
  #afsnt_arp <- rbind(afsnt_d, afsnt_a)
  afsnt_d <- afsnt_d %>% select(ARP,STT,DLY,DRR) %>%
    mutate(C02=ifelse(DRR=="C02",1,0),
           DLY_=ifelse(DLY=="Y",1,0))
  afsnt_d$schedule<-strptime(afsnt_d$STT,"%H:%M")
  afsnt_d$timeslot<-afsnt_d$schedule$hour
  afsnt_d<-afsnt_d%>%
    select(timeslot,ARP,DLY_,C02)%>%
    group_by(timeslot)%>%
    summarise(n=n(),DLY_sum=sum(DLY_),DRR_C02=sum(C02))%>%
    mutate(prop_C02=DRR_C02/n,prop_DLY=DLY_sum/n,
           ARP=paste0('ARP',i))
  return(afsnt_d)
}

D_Prop_C02 <- arp_tuner_d(afsnt,1)
for (i in 2:15){
  D_Prop_C02 <- rbind(D_Prop_C02, arp_tuner_d(afsnt,i))
}

# write.csv(D_Prop_C02, 'Proportion_C02_D.csv', row.names=F)

## AOD=A일 때
arp_tuner_a <- function(df,i){
  #afsnt_d <- df %>% filter(AOD=='D' & ARP==paste0('ARP',i))
  afsnt_a <- df %>% filter(AOD=='A' & ARP==paste0('ARP',i))
  #afsnt_arp <- rbind(afsnt_d, afsnt_a)
  afsnt_a <- afsnt_a %>% select(ARP,STT,DLY,DRR) %>%
    mutate(C02=ifelse(DRR=="C02",1,0),
           DLY_=ifelse(DLY=="Y",1,0))
  afsnt_a$schedule<-strptime(afsnt_a$STT,"%H:%M")
  afsnt_a$timeslot<-afsnt_a$schedule$hour
  afsnt_a<-afsnt_a%>%
    select(timeslot,ARP,DLY_,C02)%>%
    group_by(timeslot)%>%
    summarise(n=n(),DLY_sum=sum(DLY_),DRR_C02=sum(C02))%>%
    mutate(prop_C02=DRR_C02/n,ARP=paste0('ARP',i))
  return(afsnt_a)
}

A_Prop_C02 <- arp_tuner_a(afsnt,1)
for (i in 2:15){
  A_Prop_C02 <- rbind(A_Prop_C02, arp_tuner_a(afsnt,i))
}

# write.csv(A_Prop_C02, 'Proportion_C02_A.csv', row.names=F)

third <- AFSNT %>% group_by(REG) %>% mutate(ALL=n()) %>%
  filter(str_detect(DRR,'C')==T, DRR!='C02') %>%  
  group_by(REG) %>% mutate(N=n(),RATE=N/ALL)
b <- third %>% select(REG, RATE) %>% unique()
p <- qplot(RATE, data=b, geom='density')+
  ggtitle("기체별 불량률 분포")+
  labs(x="불량률")+
  theme(legend.title = element_blank(),plot.title = element_text(size = 20,hjust=0.5),
        legend.position = "none")
ggplotly(p)

past_tuner <- function(df){
  df <- df %>% mutate(DATE=as.Date(paste0(SDT_YY,'-',SDT_MM,'-',SDT_DD)),
                      TIME=as.numeric(gsub(':','',STT))) %>%
    group_by(DATE, REG) %>% mutate(ORDER=rank(TIME)) %>% arrange(DATE, REG, TIME)
  order <- df$ORDER; dly <- df$DLY; reg <- df$REG
  n <- nrow(df); empty <- c()
  for (i in 1:n){
    if (order[i]==1|reg[i]==''){
      empty <- empty %>% append(0)
    }
    else if (order[i]-1==order[i-1]){
      if (dly[i-1]=='Y'){
        empty <- empty %>% append(1)
      }
      else{
        empty <- empty %>% append(0)
      }
    }
    else {
      empty <- empty %>% append(NA)
    }
  }
  output <- cbind(df, PAST=empty) %>% as.data.frame() %>% select(-c(DATE, TIME, ORDER))
  return(output)
}

plt_arp<-ggplot(data=AFSNT %>% 
                  mutate(DLY=ifelse(DLY=="Y",1,0),
                         DLY_A=ifelse(str_detect(DRR,"A"),1,0)) %>% 
                  group_by(SDT_MM) %>% 
                  summarise(a=sum(DLY_A),b=n()) %>% 
                  mutate(p=a/b),aes(x=as.factor(SDT_MM),y=p))+
  geom_col(aes(fill=SDT_MM),alpha=0.6)+
  ggtitle("월별 날씨 지연율")+
  xlab("월")+
  ylab("지연율")+
  theme(plot.title = element_text(size = 20,hjust=0.5),legend.title = element_blank(),
        legend.position = "none")
weather_arp<-fread("./data/weather_arp.csv",header = T,stringsAsFactors = F)
plt_weather<-ggplot(data=weather_arp %>% 
                      group_by(SDT_MM) %>% 
                      summarise(a=mean(RN_SUM)),aes(x=as.factor(SDT_MM),y=a))+
  geom_col(aes(fill=SDT_MM),alpha=0.6)+
  ggtitle("월별 평균 강우량")+
  xlab("월")+
  ylab("평균 강우량")+
  theme(legend.title = element_blank(),plot.title = element_text(size = 20,hjust=0.5),
        legend.position = "none")
gridExtra::grid.arrange(plt_arp,plt_weather,nrow=1,ncol=2)

AFSNT <- fread("./data/AFSNT.csv",stringsAsFactors = F,header=T);afsnt<-AFSNT;
weather_arp<-fread("./data/weather_arp.csv",header = T)
prop_C02_A<-fread("./data/Proportion_C02_A.csv",header = T,stringsAsFactors = F)
prop_C02_D<-fread("./data/Proportion_C02_D.csv",header = T,stringsAsFactors = F)
airport <- fread("./data/airport_data.csv",header = T,stringsAsFactors = F)

## Schedule Complexity
arp_tuner <- function(df,i){
  afsnt_d <- df %>% filter(AOD=='D' & ARP==paste0('ARP',i))
  afsnt_a <- df %>% filter(AOD=='A' & ARP==paste0('ARP',i))
  afsnt_arp <- rbind(afsnt_d, afsnt_a)
  afsnt_arp <- afsnt_arp %>% select(SDT_YY,SDT_MM,SDT_DD,STT,ARP,ODP,AOD,FLT) %>%
    mutate(STTT=paste(SDT_YY,SDT_MM,SDT_DD,STT,sep=':'),
           POSIX=as.POSIXct(STTT,format='%Y:%m:%d:%H:%M')) %>% arrange(POSIX)
  afsnt_arp$schedule <- strptime(afsnt_arp$STTT,'%Y:%m:%d:%H:%M')
  schedule_time <- afsnt_arp$schedule$hour*60 + afsnt_arp$schedule$min
  COMPLEXITY <- c(1); tmp <- 1; cnt <- 0
  for (i in 2:nrow(afsnt_arp)){
    if(afsnt_arp$schedule[i]$mday!=afsnt_arp$schedule[i-1]$mday){
      tmp <- i; COMPLEXITY[i] <- 1
    } else{
      cnt <- 0
      for (j in tmp:i){
        if(schedule_time[i] - schedule_time[j] <= 60){
          cnt <- cnt+1
        }
      }
      COMPLEXITY[i] <- cnt
    }
  }
  afsnt_arp <- cbind(afsnt_arp, COMPLEXITY) %>% select(-c('POSIX','STTT','schedule'))
  return(afsnt_arp)
}

COMPLEXITY <- arp_tuner(afsnt,1)
for (i in 2:15){
  COMPLEXITY <- rbind(COMPLEXITY, arp_tuner(afsnt,i))
}

afsnt<-left_join(afsnt,COMPLEXITY,by=c("SDT_YY","SDT_MM","SDT_DD","STT","ARP","ODP","AOD","FLT"))

## airport data
afsnt<-left_join(afsnt,airport,by=c("ARP","SDT_YY"))

## devide AOD = A or D
afsnt_D<-afsnt %>% 
  filter(AOD=="D")

afsnt_A<-afsnt %>% 
  filter(AOD=="A")

## prop_B,C,D if AOD=A
afsnt_A$schedule<-strptime(afsnt_A$STT,"%H:%M")
afsnt_A$timeslot<-afsnt_A$schedule$hour
afsnt_A<-left_join(afsnt_A %>% select(-schedule),prop_C02_A,by=c("ARP","timeslot")) %>% 
  select(-c(timeslot,n,DLY_sum,DRR_C02)) %>% 
  dplyr::rename(C02=prop_C02)

## prop_B,C,D if AOD=D
afsnt_D$schedule<-strptime(afsnt_D$STT,"%H:%M")
afsnt_D$timeslot<-afsnt_D$schedule$hour
afsnt_D<-left_join(afsnt_D %>% select(-schedule),prop_C02_D,by=c("ARP","timeslot")) %>% 
  select(-c(timeslot,n,DLY_sum,DRR_C02,prop_DLY)) %>% 
  dplyr::rename(C02=prop_C02)


## weather data
afsnt_D$ARP_fake<-ifelse(afsnt_D$ARP %in% c("ARP2","ARP4","ARP11"),"ARP5",
                         ifelse(afsnt_D$ARP %in% c("ARP8","ARP12"),"ARP9",
                                ifelse(afsnt_D$ARP=="ARP6","ARP15",
                                       ifelse(afsnt_D$ARP=="ARP13","ARP7",
                                              ifelse(afsnt_D$ARP=="ARP14","ARP10",afsnt_D$ARP)))))
afsnt_A$ODP_fake<-ifelse(afsnt_A$ODP %in% c("ARP2","ARP4","ARP11"),"ARP5",
                         ifelse(afsnt_A$ODP %in% c("ARP8","ARP12"),"ARP9",
                                ifelse(afsnt_A$ODP=="ARP6","ARP15",
                                       ifelse(afsnt_A$ODP=="ARP13","ARP7",
                                              ifelse(afsnt_A$ODP=="ARP14","ARP10",afsnt_A$ODP)))))

afsnt_D<-left_join(afsnt_D,weather_arp,by=c("SDT_YY"="SDT_YY","SDT_MM"="SDT_MM",
                                            "ARP_fake"="ARP","SDT_DD"="DATE")) %>% 
  select(-ARP_fake)


afsnt_A<-left_join(afsnt_A,weather_arp,by=c("SDT_YY"="SDT_YY","SDT_MM"="SDT_MM",
                                            "ODP_fake"="ARP","SDT_DD"="DATE")) %>% 
  select(-ODP_fake)

afsnt<-rbind(afsnt_A,afsnt_D) %>% 
  filter(SDT_MM==9|SDT_MM==10) %>% 
  mutate(special_fit=ifelse(str_sub(FLT,-1) %in% c("T","M","F"),1,0)) %>% 
  filter(special_fit==0) %>% 
  select(-special_fit) %>% 
  filter(CNL=="N")

arp<-dummyVars(~ARP,data = afsnt,levelsOnly = T)
odp<-dummyVars(~ODP,data=afsnt,levelsOnly = T)
sdt_dy<-dummyVars(~SDT_DY,data=afsnt,levelsOnly = T)
flo<-dummyVars(~FLO,data=afsnt,levelsOnly = T)
afsnt<-cbind(afsnt,predict(arp,newdata=afsnt),predict(odp,newdata=afsnt),predict(sdt_dy,newdata=afsnt),predict(flo,newdata=afsnt))

# Train Data - PAST DELAY 입히기
past_tuner <- function(df){
  df <- df %>% mutate(DATE=as.Date(paste0(SDT_YY,'-',SDT_MM,'-',SDT_DD)),
                      TIME=as.numeric(gsub(':','',STT))) %>%
    group_by(DATE, REG) %>% mutate(ORDER=rank(TIME)) %>% arrange(DATE, REG, TIME)
  order <- df$ORDER; dly <- df$DLY; reg <- df$REG
  n <- nrow(df); empty <- c()
  for (i in 1:n){
    if (order[i]==1|reg[i]==''){
      empty <- empty %>% append(0)
    }
    else if (order[i]-1==order[i-1]){
      if (dly[i-1]=='Y'){
        empty <- empty %>% append(1)
      }
      else{
        empty <- empty %>% append(0)
      }
    }
    else {
      empty <- empty %>% append(NA)
    }
  }
  output <- cbind(df, PAST=empty) %>% as.data.frame() %>% select(-c(DATE, TIME, ORDER))
  return(output)
}

afsnt<-past_tuner(afsnt)
afsnt$PAST<-ifelse(is.na(afsnt$PAST),0,afsnt$PAST)

third <- AFSNT %>% group_by(REG) %>% mutate(ALL=n()) %>%
  filter(str_detect(DRR,'C')==T, DRR!='C02') %>%  
  group_by(REG) %>% mutate(N=n(),RATE=N/ALL)
b <- third %>% select(REG, RATE) %>% unique()
## 0.02 0.04 0.1 etc
# Left Skewed 된 분포를 띤다. 기체 당 정비 문제가 평균적으로 균일하게 일어나지만, 그 중에 예외로 정비 비율이 높은 기체가 존재한다.
REG1<-b$REG[b$RATE<=0.01]
REG2<-b$REG[b$RATE>0.01 & b$RATE <=0.04]
REG3<-b$REG[b$RATE>0.04 ]

afsnt$REG_RATE<-ifelse(afsnt$REG %in% REG1,"REG1",
                       ifelse(afsnt$REG %in% REG2,"REG2",
                              ifelse(afsnt$REG %in% REG3,"REG3",NA)))
reg_rate<-dummyVars(~REG_RATE,data=afsnt,levelsOnly = T)
nreg<-which(is.na(afsnt$REG_RATE))
afsnt<-cbind(afsnt,predict(reg_rate,newdata=afsnt))
for(i in nrow(nreg)){
  afsnt$REG_RATE[i]<-0
  afsnt$REG_RATE1[i]<-0
  afsnt$REG_RATE2[i]<-0
  afsnt$REG_RATE3[i]<-0
}

train_tuner <- function(df){
  df$INDEX <- 0
  df <- df %>% select(COMPLEXITY,APRON_SIZE,CRAFT_STAND,C02,
                      PS_AVG,WSPD_AVG,WSPD_MAX,WSPD_INS,CA_TOT_AVG,
                      RN_SUM,VIS_MNM,CLA,BASE,MIST,HM_INV,
                      ARPARP1,ARPARP2,ARPARP3,ARPARP4,ARPARP5,ARPARP6,
                      ARPARP7,ARPARP8,ARPARP9,ARPARP11,ARPARP12,
                      ARPARP13,ARPARP14,ARPARP15,
                      ODPARP1,ODPARP2,ODPARP3,ODPARP4,ODPARP5,ODPARP6,
                      ODPARP7,ODPARP8,ODPARP9,ODPARP11,ODPARP12,
                      ODPARP13,ODPARP14,ODPARP15,
                      SDT_DY월,SDT_DY화,SDT_DY수,SDT_DY목,SDT_DY금,
                      SDT_DY토,SDT_DY일,
                      FLOA,FLOB,FLOF,FLOH,FLOI,FLOJ,FLOL,
                      PAST,REG_RATEREG1,REG_RATEREG2,REG_RATEREG3,
                      INDEX,DLY,REG,SDT_DD,STT)
  return(df)
}

colnames(train_tuner(afsnt))

# write.csv(afsnt,"train_model.csv",row.names=F)

### ARP15(incheon)
dat_list_sep<-list()
for(i in 1:18){
  wd<-"./data/forecast/ARP15/"
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKSI.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP15",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKSI.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP15",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp15<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp15_ts<-list()
weather_forecast_arp15_arima<-list()
weather_forecast_arp15_pred<-list()
weather_forecast_arp15_predict<-data.frame(ARP="ARP15",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp15_ts[[i]]<-ts(weather_forecast_arp15[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp15_arima[[i]]<-arima(weather_forecast_arp15_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp15_arima[[i]],n.ahead=15)
  weather_forecast_arp15_pred[[i]]<-pred$pred
  weather_forecast_arp15_predict<-cbind(weather_forecast_arp15_predict,as.numeric(weather_forecast_arp15_pred[[i]]))
}
colnames(weather_forecast_arp15_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")
## time seires plot
ts.plot(weather_forecast_arp15_ts[[1]],weather_forecast_arp15_pred[[1]],lty=c(1,3),col="darkslategray3",main="Time-Series plot of PS_AVG")

## 최대한 수렴하는 값을 없애려고 16~30일로 제한

### ARP3(jeju)
dat_list_sep<-list()
for(i in 1:18){
  wd<-"./data/forecast/ARP3/"
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKPC.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP3",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKPC.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP3",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp3<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp3_ts<-list()
weather_forecast_arp3_arima<-list()
weather_forecast_arp3_pred<-list()
weather_forecast_arp3_predict<-data.frame(ARP="ARP3",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp3_ts[[i]]<-ts(weather_forecast_arp3[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp3_arima[[i]]<-arima(weather_forecast_arp3_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp3_arima[[i]],n.ahead=15)
  weather_forecast_arp3_pred[[i]]<-pred$pred
  weather_forecast_arp3_predict<-cbind(weather_forecast_arp3_predict,as.numeric(weather_forecast_arp3_pred[[i]]))
  # ts.plot(weather_forecast_arp15_ts[[i]],pred$pred,lty=c(1,3))
}
colnames(weather_forecast_arp3_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")


### ARP1(kimpo)
dat_list_sep<-list()
for(i in 1:18){
  wd<-"./data/forecast/ARP1/"
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKSS.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP1",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKSS.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP1",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp1<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp1_ts<-list()
weather_forecast_arp1_arima<-list()
weather_forecast_arp1_pred<-list()
weather_forecast_arp1_predict<-data.frame(ARP="ARP1",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp1_ts[[i]]<-ts(weather_forecast_arp15[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp1_arima[[i]]<-arima(weather_forecast_arp1_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp1_arima[[i]],n.ahead=15)
  weather_forecast_arp1_pred[[i]]<-pred$pred
  weather_forecast_arp1_predict<-cbind(weather_forecast_arp1_predict,as.numeric(weather_forecast_arp1_pred[[i]]))
  # ts.plot(weather_forecast_arp15_ts[[i]],pred$pred,lty=c(1,3))
}
colnames(weather_forecast_arp1_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")

### ARP7(muan)
dat_list_sep<-list()
for(i in 8:18){
  wd<-"./data/forecast/ARP7/"
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKJB.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP7",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKJB.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP7",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp7<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp7_ts<-list()
weather_forecast_arp7_arima<-list()
weather_forecast_arp7_pred<-list()
weather_forecast_arp7_predict<-data.frame(ARP="ARP7",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp7_ts[[i]]<-ts(weather_forecast_arp7[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp7_arima[[i]]<-arima(weather_forecast_arp15_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp7_arima[[i]],n.ahead=15)
  weather_forecast_arp7_pred[[i]]<-pred$pred
  weather_forecast_arp7_predict<-cbind(weather_forecast_arp7_predict,as.numeric(weather_forecast_arp7_pred[[i]]))
  # ts.plot(weather_forecast_arp15_ts[[i]],pred$pred,lty=c(1,3))
}
colnames(weather_forecast_arp7_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")

### ARP5(ulsan)
dat_list_sep<-list()
for(i in 1:18){
  wd<-"./data/forecast/ARP5/"
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKPU.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP5",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKPU.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP5",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp5<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp5_ts<-list()
weather_forecast_arp5_arima<-list()
weather_forecast_arp5_pred<-list()
weather_forecast_arp5_predict<-data.frame(ARP="ARP5",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp5_ts[[i]]<-ts(weather_forecast_arp5[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp5_arima[[i]]<-arima(weather_forecast_arp5_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp5_arima[[i]],n.ahead=15)
  weather_forecast_arp5_pred[[i]]<-pred$pred
  weather_forecast_arp5_predict<-cbind(weather_forecast_arp5_predict,as.numeric(weather_forecast_arp5_pred[[i]]))
  # ts.plot(weather_forecast_arp15_ts[[i]],pred$pred,lty=c(1,3))
}
colnames(weather_forecast_arp5_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")

### ARP10(yang)
dat_list_sep<-list()
for(i in 2:18){
  wd<-"./data/forecast/ARP10/"
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKNY.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP10",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKNY.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP10",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp10<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp10_ts<-list()
weather_forecast_arp10_arima<-list()
weather_forecast_arp10_pred<-list()
weather_forecast_arp10_predict<-data.frame(ARP="ARP10",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp10_ts[[i]]<-ts(weather_forecast_arp10[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp10_arima[[i]]<-arima(weather_forecast_arp10_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp10_arima[[i]],n.ahead=15)
  weather_forecast_arp10_pred[[i]]<-pred$pred
  weather_forecast_arp10_predict<-cbind(weather_forecast_arp10_predict,as.numeric(weather_forecast_arp10_pred[[i]]))
  # ts.plot(weather_forecast_arp15_ts[[i]],pred$pred,lty=c(1,3))
}
colnames(weather_forecast_arp10_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")

### ARP9(yusu)
dat_list_sep<-list()
for(i in 1:18){
  wd<-c("./data/forecast/ARP9/")
  if(i<10){
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_200",i,"09_RKJY.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP9",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  } else {
    dat_list_sep[[i]]<-read.csv(paste0(wd,"MONTH_20",i,"09_RKJY.csv"),header = T,fileEncoding = "UTF8",stringsAsFactors = F)
    dat_list_sep[[i]]<-cbind((2000+i),9,"ARP9",dat_list_sep[[i]])
    colnames(dat_list_sep[[i]])[c(1,2,3)]<-c("SDT_YY","SDT_MM","ARP")
    dat_list_sep[[i]]$RN_SUM<-ifelse(is.na(dat_list_sep[[i]]$RN_SUM),0,dat_list_sep[[i]]$RN_SUM)
    dat_list_sep[[i]]$SD_MAX<-ifelse(is.na(dat_list_sep[[i]]$SD_MAX),0,dat_list_sep[[i]]$SD_MAX)
    dat_list_sep[[i]]$VIS_MNM<-ifelse(is.na(dat_list_sep[[i]]$VIS_MNM),1000,dat_list_sep[[i]]$VIS_MNM)
    dat_list_sep[[i]]$CLA<-ifelse(is.na(dat_list_sep[[i]]$CLA),0,dat_list_sep[[i]]$CLA)
    dat_list_sep[[i]]$BASE<-ifelse(is.na(dat_list_sep[[i]]$BASE),0,dat_list_sep[[i]]$BASE)
  }
}
weather_forecast_arp9<-rbind(dat_list_sep[[1]],dat_list_sep[[2]],dat_list_sep[[3]],dat_list_sep[[4]],dat_list_sep[[5]],dat_list_sep[[6]],dat_list_sep[[7]],dat_list_sep[[8]],dat_list_sep[[9]],dat_list_sep[[10]],dat_list_sep[[11]],dat_list_sep[[12]],dat_list_sep[[13]],dat_list_sep[[14]],dat_list_sep[[15]],dat_list_sep[[16]],dat_list_sep[[17]],dat_list_sep[[18]]) %>% 
  filter(DATE>=16) %>% 
  select(-c(CLF,SD_MAX,WD_MAX,WD_INS))
weather_forecast_arp9_ts<-list()
weather_forecast_arp9_arima<-list()
weather_forecast_arp9_pred<-list()
weather_forecast_arp9_predict<-data.frame(ARP="ARP9",DATE=c(16:30))
# par(mfrow=c(5,5))
for(i in c("PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")){
  weather_forecast_arp9_ts[[i]]<-ts(weather_forecast_arp9[,i],start = c(2001,1),frequency = 15)
  weather_forecast_arp9_arima[[i]]<-arima(weather_forecast_arp9_ts[[i]],c(2,0,0))
  pred<-predict(weather_forecast_arp9_arima[[i]],n.ahead=15)
  weather_forecast_arp9_pred[[i]]<-pred$pred
  weather_forecast_arp9_predict<-cbind(weather_forecast_arp9_predict,as.numeric(weather_forecast_arp9_pred[[i]]))
  # ts.plot(weather_forecast_arp15_ts[[i]],pred$pred,lty=c(1,3))
}
colnames(weather_forecast_arp9_predict)<-c("ARP","DATE","PS_AVG","TMP_AVG","TMP_MAX","TMP_MNM","TD_AVG","HM_AVG","HM_MNM","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE")

## export the data
# write.csv(weather_forecast_arp1,"weather_forecast_arp1.csv",row.names = F)
# write.csv(weather_forecast_arp3,"weather_forecast_arp3.csv",row.names = F)
# write.csv(weather_forecast_arp5,"weather_forecast_arp5.csv",row.names = F)
# write.csv(weather_forecast_arp7,"weather_forecast_arp7.csv",row.names = F)
# write.csv(weather_forecast_arp9,"weather_forecast_arp9.csv",row.names = F)
# write.csv(weather_forecast_arp10,"weather_forecast_arp10.csv",row.names = F)
# write.csv(weather_forecast_arp15,"weather_forecast_arp15.csv",row.names = F)

weather_forecast_predict<-rbind(weather_forecast_arp1_predict,weather_forecast_arp3_predict,weather_forecast_arp5_predict,weather_forecast_arp7_predict,weather_forecast_arp9_predict,weather_forecast_arp10_predict,weather_forecast_arp15_predict) %>% 
  mutate(MIST=TMP_MNM-TD_AVG,HM_INV=1/HM_AVG) %>% 
  select(-c(TMP_AVG,TMP_MAX,TMP_MNM,TD_AVG,HM_AVG,HM_MNM))
colnames(weather_forecast_predict)<-c("ARP","DATE","PS_AVG","WSPD_AVG","WSPD_MAX","WSPD_INS","CA_TOT_AVG","RN_SUM","VIS_MNM","CLA","BASE","MIST","HM_INV")
# write.csv(weather_forecast_predict,"weather_forecast_predict.csv",row.names=F)

AFSNT<-fread("./data/AFSNT.csv",header = T,stringsAsFactors = F); afsnt<-AFSNT
AFSNT_DLY<-fread("./data/AFSNT_DLY.csv",header = T,stringsAsFactors = F)
AFSNT_DLY[3,5]<-"ARP1"
weather_forecast<-fread("./data/weather_forecast_predict.csv",header = T,stringsAsFactors = F)
prop_C02_A<-fread("./data/Proportion_C02_A.csv",header = T,stringsAsFactors = F)
prop_C02_D<-fread("./data/Proportion_C02_D.csv",header = T,stringsAsFactors = F)
airport <- fread("./data/airport_data.csv",header = T,stringsAsFactors = F)


afsnt <- afsnt %>% mutate(DATE=as_date(paste(SDT_YY,'-',SDT_MM,'-',SDT_DD)),
                          SCT=as.integer(gsub(':','',STT)), # 순서를 매기기 위해
                          STT2=hm(STT), # 시-분 형태로 표현
                          ATT2=hm(ATT), # 시-분 형태로 표현
                          DIF=ATT2-STT2) # 지연된 시간
afsnt$DIF <- afsnt$DIF@hour*60 + afsnt$DIF@minute # 분으로 표현
afsnt <- afsnt %>% select(-c(STT2, ATT2))
###afsnt DLY 시간 날짜 정리

AFSNT_DLY <- AFSNT_DLY %>% mutate(DATE=as_date(paste(SDT_YY,'-',SDT_MM,'-',SDT_DD)),
                                  SCT=as.integer(gsub(':','',STT)),
                                  INDEX=1:nrow(AFSNT_DLY))

#######################
# SFSNT 와 AFSNT 를 통한 REG 의 루틴을 Test data 에 적용 

#######train data  REG  - Test data 에 입히기
#최신 train data 일정 뽑기 - 2019년 6월 마지막주의 FLT를 통하여..
AFSNT_DLY$DY_FLT <- paste0(AFSNT_DLY$SDT_DY,AFSNT_DLY$FLT,AFSNT_DLY$AOD)

latest_afsnt <- afsnt %>% filter(DATE>='2019-06-24')
latest_afsnt$DY_FLT <- paste0(latest_afsnt$SDT_DY,latest_afsnt$FLT,latest_afsnt$AOD) 

latest_afsnt <- latest_afsnt %>% select(DY_FLT,REG) 

#AFSNT 와 최신 일정 을 JOIN 한 , AFSNT_DLY_join 이라는 새로운 데이터 프레임 생성
AFSNT_DLY_join <- left_join(AFSNT_DLY,latest_afsnt, by= 'DY_FLT') %>% arrange(INDEX)
AFSNT_DLY_join$REG[is.na(AFSNT_DLY_join$REG)] <- ""

#### (최신 항공편) 6월 마지막 주에 없는 항공편 채워 넣기
#  REG가 정해지지 않은 항공편만 뽑아내기 
NAREG <- AFSNT_DLY_join %>%  filter(AFSNT_DLY_join$REG=="")
NAREG <- NAREG[,1:16] # reg 컴럼 삭제 

#########
#REG가 정해지지 않은 FLT들만 뽑아 내어  6월 셋쨰 주 운항 일정 입히기 
#셋쨰 주 선택 

W3_afsnt <- afsnt %>% filter(DATE>='2019-06-16',DATE<='2019-06-22')
W3_afsnt$DY_FLT <- paste0(W3_afsnt$SDT_DY,W3_afsnt$FLT,W3_afsnt$AOD) 
W3_afsnt <- W3_afsnt %>% select(REG, DY_FLT)

#NA REG와 합치기 
NA_REG_join <- left_join(NAREG,W3_afsnt, by= 'DY_FLT') %>% arrange(INDEX)

NA_REG_join$DATE_FLT_AOD <- paste0(NA_REG_join$DATE,NA_REG_join$FLT,NA_REG_join$AOD)
NA_REG_join <-NA_REG_join %>% select(DATE_FLT_AOD,REG)

AFSNT_DLY_join$DATE_FLT_AOD <- paste0(AFSNT_DLY_join$DATE,AFSNT_DLY_join$FLT,AFSNT_DLY_join$AOD)

# 셋쨰주 운항 일정으로 새롭게 찾은 정보들을 원본파일에 합치기 
AFSNT_DLY_join <- left_join(AFSNT_DLY_join,NA_REG_join,by='DATE_FLT_AOD')

AFSNT_DLY_join$REG.x[is.na(AFSNT_DLY_join$REG.x)] <- ""
AFSNT_DLY_join$REG.y[is.na(AFSNT_DLY_join$REG.y)] <- ""

AFSNT_DLY_join$REG.x<-ifelse(AFSNT_DLY_join$REG.x=="",AFSNT_DLY_join$REG.y,AFSNT_DLY_join$REG.x)

AFSNT_DLY_join<- AFSNT_DLY_join[,-c(18,19)] # reg y  & 불필요 칼럼 삭제
names(AFSNT_DLY_join)[names(AFSNT_DLY_join) == "REG.x"] <- c("REG") 

###########
###### 같은 FLT 에 다수의 REG 가 있는 경우는 확실한 정보가 아니므로 공백 처리 
AFSNT_DLY_join$DATE_DY_FLT <- paste0(AFSNT_DLY_join$DATE,AFSNT_DLY_join$DY_FLT)
AFSNT_DLY_join$idx <-  1:nrow(AFSNT_DLY_join)


dup_del <-AFSNT_DLY_join[-which(duplicated(AFSNT_DLY_join$DATE_DY_FLT)),]
anti_del <- anti_join(AFSNT_DLY_join,dup_del, by = 'idx')
anti_del_idx <- anti_join(AFSNT_DLY_join,dup_del, by = 'idx') %>% select(idx)

dup_del$REG <- ifelse(dup_del$idx %in% anti_del_idx, dup_del$REG=="",dup_del$REG)

dup_del<-dup_del %>% select(-c(DY_FLT,DATE_DY_FLT,idx,DATE,SCT))

afsnt_dly<-dup_del

## complex
  arp_tuner <- function(df,i){
    afsnt_d <- df %>% filter(AOD=='D' & ARP==paste0('ARP',i))
    afsnt_a <- df %>% filter(AOD=='A' & ARP==paste0('ARP',i))
    afsnt_arp <- rbind(afsnt_d, afsnt_a)
    afsnt_arp <- afsnt_arp %>% select(SDT_YY,SDT_MM,SDT_DD,STT,ARP,ODP,AOD,FLT) %>%
      mutate(STTT=paste(SDT_YY,SDT_MM,SDT_DD,STT,sep=':'),
             POSIX=as.POSIXct(STTT,format='%Y:%m:%d:%H:%M')) %>% arrange(POSIX)
    afsnt_arp$schedule <- strptime(afsnt_arp$STTT,'%Y:%m:%d:%H:%M')
    schedule_time <- afsnt_arp$schedule$hour*60 + afsnt_arp$schedule$min
    COMPLEXITY <- c(1); tmp <- 1; cnt <- 0
    for (i in 2:nrow(afsnt_arp)){
      if(afsnt_arp$schedule[i]$mday!=afsnt_arp$schedule[i-1]$mday){
        tmp <- i; COMPLEXITY[i] <- 1
      } else{
        cnt <- 0
        for (j in tmp:i){
          if(schedule_time[i] - schedule_time[j] <= 60){
            cnt <- cnt+1
          }
        }
        COMPLEXITY[i] <- cnt
      }
    }
    afsnt_arp <- cbind(afsnt_arp, COMPLEXITY) %>% select(-c('POSIX','STTT','schedule'))
    return(afsnt_arp)
  }
  
  COMPLEXITY <- arp_tuner(afsnt_dly,1)
  for (i in 2:15){
    COMPLEXITY <- rbind(COMPLEXITY, arp_tuner(afsnt_dly,i))
  }
  
  afsnt_dly<-left_join(afsnt_dly,COMPLEXITY,by=c("SDT_YY","SDT_MM","SDT_DD","STT","ARP","ODP","AOD","FLT"))

## airport data
afsnt_dly<-left_join(afsnt_dly,airport,by=c("ARP","SDT_YY"))

## devide AOD = A or D
afsnt_dly_D<-afsnt_dly %>% filter(AOD=="D")
afsnt_dly_A<-afsnt_dly %>% filter(AOD=="A")

## prop C02, AOD=A
afsnt_dly_A$schedule<-strptime(afsnt_dly_A$STT,"%H:%M")
afsnt_dly_A$timeslot<-afsnt_dly_A$schedule$hour
afsnt_dly_A<-left_join(afsnt_dly_A %>% select(-schedule), prop_C02_A ,by=c("ARP","timeslot"))

afsnt_dly_A <- afsnt_dly_A %>% select(-c(timeslot,n,DLY_sum,DRR_C02)) %>%dplyr::rename(C02=prop_C02)

afsnt_dly_A$C02<-ifelse(is.na(afsnt_dly_A$C02),0,afsnt_dly_A$C02)

## prop C02, AOD=D
afsnt_dly_D$schedule<-strptime(afsnt_dly_D$STT,"%H:%M")
afsnt_dly_D$timeslot<-afsnt_dly_D$schedule$hour
afsnt_dly_D<-left_join(afsnt_dly_D %>% select(-schedule),prop_C02_D,by=c("ARP","timeslot")) %>% 
  select(-c(timeslot,n,DLY_sum,DRR_C02,prop_DLY)) %>% 
  dplyr::rename(C02=prop_C02)
afsnt_dly_D$C02<-ifelse(is.na(afsnt_dly_D$C02),0,afsnt_dly_D$C02)

afsnt_dly_D$ARP_fake <- ifelse(afsnt_dly_D$ARP %in% c("ARP2","ARP4","ARP11"),"ARP5",
                         ifelse(afsnt_dly_D$ARP %in% c("ARP8","ARP12"),"ARP9",
                                ifelse(afsnt_dly_D$ARP=="ARP6","ARP15",
                                       ifelse(afsnt_dly_D$ARP=="ARP13","ARP7",
                                              ifelse(afsnt_dly_D$ARP=="ARP14","ARP10",afsnt_dly_D$ARP)))))

afsnt_dly_A$ODP_fake<-ifelse(afsnt_dly_A$ODP %in% c("ARP2","ARP4","ARP11"),"ARP5",
                         ifelse(afsnt_dly_A$ODP %in% c("ARP8","ARP12"),"ARP9",
                                ifelse(afsnt_dly_A$ODP=="ARP6","ARP15",
                                       ifelse(afsnt_dly_A$ODP=="ARP13","ARP7",
                                              ifelse(afsnt_dly_A$ODP=="ARP14","ARP10",afsnt_dly_A$ODP)))))

afsnt_dly_D <- left_join(afsnt_dly_D,weather_forecast,by=c("ARP_fake"="ARP","SDT_DD"="DATE")) %>% select(-ARP_fake)
afsnt_dly_A<-left_join(afsnt_dly_A,weather_forecast,by=c("ODP_fake"="ARP","SDT_DD"="DATE")) %>% select(-ODP_fake)
afsnt_dly<-rbind(afsnt_dly_A,afsnt_dly_D)

arp<-dummyVars(~ARP,data = afsnt_dly,levelsOnly = T)
odp<-dummyVars(~ODP,data=afsnt_dly,levelsOnly = T)
sdt_dy<-dummyVars(~SDT_DY,data=afsnt_dly,levelsOnly = T)
flo<-dummyVars(~FLO,data=afsnt_dly,levelsOnly = T)
afsnt_dly<-cbind(afsnt_dly,predict(arp,newdata=afsnt_dly),predict(odp,newdata=afsnt_dly),predict(sdt_dy,newdata=afsnt_dly),predict(flo,newdata=afsnt_dly))
afsnt_dly$PAST<-0

third <- AFSNT %>% group_by(REG) %>% mutate(ALL=n()) %>%
  filter(str_detect(DRR,'C')==T, DRR!='C02') %>%  
  group_by(REG) %>% mutate(N=n(),RATE=N/ALL)
b <- third %>% select(REG, RATE) %>% unique()
## 0.02 0.04 0.1 etc
# Left Skewed 된 분포를 띤다. 기체 당 정비 문제가 평균적으로 균일하게 일어나지만, 그 중에 예외로 정비 비율이 높은 기체가 존재한다.
REG1<-b$REG[b$RATE<=0.01]
REG2<-b$REG[b$RATE>0.01 & b$RATE <=0.04]
REG3<-b$REG[b$RATE>0.04 ]
afsnt_dly$REG_RATE<-ifelse(afsnt_dly$REG %in% REG1,"REG1",
                       ifelse(afsnt_dly$REG %in% REG2,"REG2",
                              ifelse(afsnt_dly$REG %in% REG3,"REG3",NA)))

reg_rate<-dummyVars(~REG_RATE,data=afsnt_dly,levelsOnly = T)
afsnt_dly<-cbind(afsnt_dly,predict(reg_rate,newdata=afsnt_dly))

nreg<-which(is.na(afsnt_dly$REG_RATE))
for(i in nreg){
  afsnt_dly$REG_RATE[i]<-0
  afsnt_dly$REG_RATEREG1[i]<-0
  afsnt_dly$REG_RATEREG2[i]<-0
  afsnt_dly$REG_RATEREG3[i]<-0
}

afsnt_dly<-afsnt_dly %>% 
  select(-c(ARPARP10,ODPARP10)) %>% 
  arrange(INDEX)

test_tuner <- function(df){
  df <- df %>% select(COMPLEXITY,APRON_SIZE,CRAFT_STAND,C02,
                      PS_AVG,WSPD_AVG,WSPD_MAX,WSPD_INS,CA_TOT_AVG,
                      RN_SUM,VIS_MNM,CLA,BASE,MIST,HM_INV,
                      ARPARP1,ARPARP2,ARPARP3,ARPARP4,ARPARP5,ARPARP6,
                      ARPARP7,ARPARP8,ARPARP9,ARPARP11,ARPARP12,
                      ARPARP13,ARPARP14,ARPARP15,
                      ODPARP1,ODPARP2,ODPARP3,ODPARP4,ODPARP5,ODPARP6,
                      ODPARP7,ODPARP8,ODPARP9,ODPARP11,ODPARP12,
                      ODPARP13,ODPARP14,ODPARP15,
                      SDT_DY월,SDT_DY화,SDT_DY수,SDT_DY목,SDT_DY금,
                      SDT_DY토,SDT_DY일,
                      FLOA,FLOB,FLOF,FLOH,FLOI,FLOJ,FLOL,
                      PAST,REG_RATEREG1,REG_RATEREG2,REG_RATEREG3,
                      INDEX,DLY,REG,SDT_DD,STT)
  return(df)
}

colnames(test_tuner(afsnt_dly) %>% select(-c(INDEX,DLY,REG,SDT_DD,STT)))

# write.csv(afsnt_dly,"afsnt_dly_predict.csv",row.names = F)

train <- fread("./data/train_model.csv",header=T,stringsAsFactors=F) %>% 
  mutate(DLY=ifelse(DLY=="Y",1,0))
test <- fread('./data/afsnt_dly_predict.csv',header=T,stringsAsFactors=F) %>% 
  mutate(DLY=ifelse(DLY=="Y",1,0))
afsnt_dly <- fread('./data/afsnt_dly.csv',header=T,stringsAsFactors=F) %>% select(-c(DLY, DLY_RATE))
afsnt_dly$INDEX <- 1:nrow(afsnt_dly)

normalize <- function(x){
  return((x - min(x))/(max(x)-min(x)))
}

train_tuner <- function(df){
  df$INDEX <- 0
  df <- df %>% select(COMPLEXITY,APRON_SIZE,CRAFT_STAND,C02,
                      PS_AVG,WSPD_AVG,WSPD_MAX,WSPD_INS,CA_TOT_AVG,
                      RN_SUM,VIS_MNM,CLA,BASE,MIST,HM_INV,
                      ARPARP1,ARPARP2,ARPARP3,ARPARP4,ARPARP5,ARPARP6,
                      ARPARP7,ARPARP8,ARPARP9,ARPARP11,ARPARP12,
                      ARPARP13,ARPARP14,ARPARP15,
                      ODPARP1,ODPARP2,ODPARP3,ODPARP4,ODPARP5,ODPARP6,
                      ODPARP7,ODPARP8,ODPARP9,ODPARP11,ODPARP12,
                      ODPARP13,ODPARP14,ODPARP15,
                      SDT_DY월,SDT_DY화,SDT_DY수,SDT_DY목,SDT_DY금,
                      SDT_DY토,SDT_DY일,
                      FLOA,FLOB,FLOF,FLOH,FLOI,FLOJ,FLOL,
                      PAST,REG_RATEREG1,REG_RATEREG2,REG_RATEREG3,
                      INDEX,DLY,REG,SDT_DD,STT)
  return(df)
}

test_tuner <- function(df){
  df <- df %>% select(COMPLEXITY,APRON_SIZE,CRAFT_STAND,C02,
                      PS_AVG,WSPD_AVG,WSPD_MAX,WSPD_INS,CA_TOT_AVG,
                      RN_SUM,VIS_MNM,CLA,BASE,MIST,HM_INV,
                      ARPARP1,ARPARP2,ARPARP3,ARPARP4,ARPARP5,ARPARP6,
                      ARPARP7,ARPARP8,ARPARP9,ARPARP11,ARPARP12,
                      ARPARP13,ARPARP14,ARPARP15,
                      ODPARP1,ODPARP2,ODPARP3,ODPARP4,ODPARP5,ODPARP6,
                      ODPARP7,ODPARP8,ODPARP9,ODPARP11,ODPARP12,
                      ODPARP13,ODPARP14,ODPARP15,
                      SDT_DY월,SDT_DY화,SDT_DY수,SDT_DY목,SDT_DY금,
                      SDT_DY토,SDT_DY일,
                      FLOA,FLOB,FLOF,FLOH,FLOI,FLOJ,FLOL,
                      PAST,REG_RATEREG1,REG_RATEREG2,REG_RATEREG3,
                      INDEX,DLY,REG,SDT_DD,STT)
  return(df)
}

tuned_train <- train_tuner(train)
tuned_test <- test_tuner(test)

tuned_train$label <- 'train'
tuned_test$label <- 'test'
pool <- rbind(tuned_train, tuned_test)

pool_norm <- cbind(as.data.frame(lapply(pool[,1:61], normalize)),
                   INDEX=pool$INDEX,
                   DLY=pool$DLY,
                   REG=pool$REG,
                   SDT_DD=pool$SDT_DD,
                   STT=pool$STT,
                   label=pool$label)

train_norm <- pool_norm %>% filter(label=='train') %>% select(-label)
test_norm <- pool_norm %>% filter(label=='test') %>% select(-label)

train <- train_norm %>% select(-c(INDEX,REG,SDT_DD,STT))

train_temp <- list()
train_actual <- list()
test_temp <- list()
table(train$DLY)

for (k in 1:5){
  train_temp[[k]] <- ovun.sample(DLY~., data=train, N=5000, method="both", p=0.3, seed=k)$data
  train_actual[[k]] <- ovun.sample(DLY~., data=train, N=10000, method="both", p=0.3, seed=k)$data
}
for (k in 1:5){
  set.seed(k)
  idx <- sample(1:nrow(train), 5000*0.3)
  test_temp[[k]] <- train[idx,]
}
table(train_temp[[k]]$DLY)

test_final <- test_norm %>% mutate(STT=as.integer(gsub(':','',STT))) %>% select(-DLY)